The present invention relates to a technique, which is particularly applicable to a read method for multiple-value information in a semiconductor memory and, moreover, is effective for a nonvolatile semiconductor memory. More particularly the invention is applicable to a technique which may be effectively used for a nonvolatile memory (hereafter referred to as a flash memory) to electrically simultaneously erase a plurality of stored pieces of information.
A flash memory uses a nonvolatile memory cell having a control gate and a floating gate as a memory cell. It is possible to constitute the memory cell of one transistor. In the case of a write operation of the flash memory, as shown in FIG. 16, a state in which the threshold voltage is low (logic xe2x80x9c0xe2x80x9d) is set up by setting the drain region of a nonvolatile memory cell at approx. 5 V (volt) and a word line connected with a control gate CG at approx. xe2x88x9211 V and, thereby, extracting electric charges from a floating gate FG by means of a tunnel current. In the case of an erase operation, as shown in FIG. 17, a state in which the threshold voltage is high (logic xe2x80x9c1xe2x80x9d) is set up by setting a well region, drain region and source region at approximately 0V and the control gate CG to a high voltage, such as 16V, thereby generating a tunnel current, and injecting negative electric charges into the floating gate CG. During the read operation, it is judged that the data stored in a memory cell through which current flows is xe2x80x9c0xe2x80x9d and a memory cell through which no current flows is xe2x80x9c1xe2x80x9d by setting the control gate at an intermediate voltage between a high threshold and a low threshold and detecting whether current flows or not. Thereby, one-bit of data is stored in one memory cell.
A technique has been proposed, which is related to the so-called multiple-value memory, for storing data of two bits or more in one memory cell in order to increase the memory capacity. An invention related to the multiple-value memory is disclosed in Japanese Patent Application No. 14031/1995, etc.
Such a multiple-value memory stores information by controlling the amount of electric charges to be injected into a floating gate, thereby stepwise changing thresholds to 1V, 2V, 3V, . . . , and making information of a plurality of bits correspond to each threshold value. FIG. 18 shows a threshold value distribution state when storing information by dividing one memory cell into four threshold value states (this will be referred to as four-value state in this specification). It is difficult to accurately control the threshold value of a memory cell to a predetermined value for a write operation, and therefore, as shown in FIG. 18, a normal distribution is established around each target threshold voltage. To read data, voltages corresponding to the valleys of the threshold value distributions are read, set as VRW1, VRW2 and VRW3, and applied to a control gate through a word line. In this case, the drain is set at a potential, such as 1V, and the source is set at a potential, such as 0V. The bit-line precharging method can be used for the setting of the drain voltage.
Table 1 shows the results of reading data from memory cells belonging to the threshold value distributions A, B, C, and D by using the above read voltages VRW1, VRW2, and VRW3 (VRW1 less than VRW2 less than VRW3). Because the memory cell belonging to the threshold value distribution A has the highest threshold value, no current flows even if any one of VRW1, VRW2, and VRW3 is applied. Therefore, the read result is xe2x80x9c1xe2x80x9d. In the case of the memory cell belonging to the threshold value distribution B, the read result is xe2x80x9c1xe2x80x9d because no current flows when VRW1 or VRW2 is applied. However, when VRW3 is applied, the read result is xe2x80x9c0xe2x80x9d because current flows. In the case of the memory cell belonging to the threshold value distribution C, the read result is xe2x80x9c1xe2x80x9d because no current flows when VRW1 is applied. However, when VRW2 or VRW3 is applied, the read result is xe2x80x9c0xe2x80x9d because current flows. In the case of the memory cell belonging to the threshold value distribution D, the read result is always xe2x80x9c0xe2x80x9d in any case because current flows if any one of VRW1, VRW2, and VRW3 is applied. Though a case of a four-value memory has been described above, the same holds theoretically for eight- and sixteen-value memories.
In the case of a four-value memory, it is possible to store two-bit information because any one of four threshold values can be set in one memory cell. In the case of a conventional two-value memory for storing one-bit of information in one memory, the read operation is performed once because two threshold values are judged to obtain one-bit of information. In the case of a four-value memory, however, it is necessary to perform the read operation three times by changing the potentials of a word line in order to obtain two-bit information. Therefore, simply saying, a problem arises that the read time is three times larger than that of a two-value memory and the power consumption for the read operation also increases three times.
Moreover, in the case of a nonvolatile memory using a MOSFET having a floating gate as a memory cell, a phenomenon (hereafter referred to as read disturb) occurs wherein a small amount of hot electrons generated in the read operation are injected into the floating gate when the read operation is repeated, and, thereby, the threshold value is raised. Therefore, a drawback occurs in that the threshold value fluctuation in a memory cell increases as the read frequency increases, exceeds the read level in the worst case and, resultingly, the stored data may be changed.
Moreover, as described above, it is necessary to apply the earthing potential Vss (0 V) to the source of a memory cell during the read operation. As shown in FIG. 19, a power supply line (ground line) GL for the purpose is extended from the external terminal (ground pin) GND of a chip to each memory cell MC in a memory array M-ARY. The power supply line is generally constituted of a metal wiring layer made of, for example, aluminum. However, at a portion where the aluminum wiring is used for another signal line, such as a bit line, the earthing potential may be applied to each memory cell MC through a diffusion layer having a large resistance value. In this case, as shown in FIG. 19, the ground line length differs between the memory cell closest to the ground GND and the memory cell farthest from the ground GND. For example, the wiring of the diffusion layer has a resistance value of hundreds of mxcexa9 per xcexcw and even a metal wiring has a resistance value of approx. 100xcexa9. Therefore, a portion between the ground pin and the memory cell farthest from the ground pin GND has hundreds to thousands of xcexa9. Therefore, when a current flows from a memory cell during a read operation, the source potential rises and, thereby, the source potential greatly differs between the memory cell closest to the ground pin and the memory cell farthest from the ground pin. If the read current is 3 mA and the ground resistance differs by 100xcexa9, a difference of 0.3 V occurs in the source potential.
In the case of a memory cell, the drain current logarithmically changes nearby the threshold value as known from the characteristics of a MOSFET. Therefore, when the source potential rises and the voltage between the gate and source lowers, the current decreases by one-tenth to one-hundredth. In this case, if the characteristics of memory cells belonging to the distributions B and D in FIG. 18 have the characteristics shown by b and d in FIG. 20, a sufficient amount of current can be ensured even if the source potential slightly rises when applying a voltage of 5 V to each gate because the memory cell D is in a completely saturated region. However, because the memory cell B is only slightly saturated, it is found that the current is greatly decreased due to a slight rise of the source potential and, thereby, no data can be read or erroneous data may be read.
It is an object of the present invention to provide a multiple-value storage nonvolatile semiconductor memory having a short read time and a small power consumption.
It is another object of the present invention to provide a multiple-value storage nonvolatile semiconductor memory which is capable of minimizing the necessary read frequency and preventing stored data from easily changing.
It is still another object of the present invention to provide a multiple-value storage nonvolatile semiconductor memory which is capable of preventing the source potential from rising during a read operation and preventing a data read operation from failing or data from being erroneously read.
The above and other objects and novel features of the present invention will become more apparent from the description provided in this specification and the accompanying drawings.
The outline of a representative one of the embodiments disclosed in this application will be briefly described below.
That is, in the case of a nonvolatile semiconductor memory where multiple-value information is stored in one memory cell by setting a plurality of threshold values, data is successively read while changing the word-line read level from a lower value to a higher one and selectively precharging a bit line for the next read operation in accordance with the data stored in latching means for holding read data.
As understood by referring to Table 1, by changing the word-line read level from a lower value to a higher one successively, xe2x80x9c0xe2x80x9d is read from a memory cell from which xe2x80x9c0xe2x80x9d has already been read even after reading data from the memory cell by raising the read level. Therefore, the result is the same as the case when the read operation is not performed. That is, it is possible to omit the precharging of the bit line. Moreover, because the current consumption can be reduced by the omission of the precharging, it is possible to reduce the current flowing from a memory array to a ground line during the read operation. Thereby, it is possible to decrease the rise of the source potential of a memory cell. Therefore, it is possible to prevent a data read operation from failing or data from being erroneously read. Moreover, because the read frequency can be decreased by the omission of precharging, it is possible to suppress the threshold value fluctuation due to a read disturb condition, that is, to suppress a change in the stored data. Furthermore, by the above read method, because all of the read data will become xe2x80x9c0xe2x80x9d before the memory cells connected to a selected word line are read when they have a low threshold value, it is possible to interrupt the read operation by using all-xe2x80x9c0xe2x80x9d judging means and, moreover, to reduce the current consumption and the data read time.
Furthermore, it is preferable to determine the correspondence between the threshold values of memory cells and the data stored in the memory cells so that the codes of the stored data are so arranged as to be different only in one bit between adjacent threshold values. Specifically, in the case of a four-value memory, the threshold value distributions A, B, C, and D of FIG. 18 are made to correspond to two-bit data values xe2x80x9c11xe2x80x9d, xe2x80x9c10xe2x80x9d, xe2x80x9c00xe2x80x9d, and xe2x80x9c01xe2x80x9d. Thus, there is an advantage that, even if a read disturb condition occurs, the load of an error correction circuit for correcting the read disturb condition can be lightened and the circuit scale also can be decreased. For example, if the threshold value of a memory cell in the threshold value distribution B of FIG. 18 is shifted to the threshold value distribution C due to a disturb condition, only one bit has an error because the original stored data xe2x80x9c10xe2x80x9d is erroneously read as xe2x80x9c00xe2x80x9d in the case of the above correspondence. However, when the threshold value distributions A, B, C, and D of FIG. 18 are made to correspond to two-bit data values xe2x80x9c11xe2x80x9d, xe2x80x9c10xe2x80x9d, xe2x80x9c01xe2x80x9d, and xe2x80x9c00xe2x80x9d, if the threshold value of the memory cell is shifted from B to C, xe2x80x9c10xe2x80x9d is erroneously read as xe2x80x9c01xe2x80x9d and, thereby, a two-bit error occurs. To correct the two-bit error, however, the load of the error correction circuit and the circuit scale are greatly increased.